wav.write("sound/" + experiment_name + "_random.wav", sample_rate, random)
wav.write("sound/" + experiment_name + "_reconstruction.wav", sample_rate,
reconstruction)
wav.write("sound/" + experiment_name + "_reconstruction_original.wav",
sample_rate, original)
ou1 = net1.output(x_train[:50])
ou2 = net2.output(x_train[:50])
for i in range(ou2.shape[0]):
if i < 100:
plt.figure()
plt.plot(ou1[i, 0, :, 0], color="g")
plt.plot(ou2[i, 0, :, 0], color="b")
plt.plot(x_train[i, 0, :, 0], color="r")
plt.savefig(plot_dir + str(i) + "_compare.png", cmap=plt.cm.binary)
pickle_saver(net2.params, "model_no_boost.pkl")
f = open("readme", "w")
f.write("SingVAE parameters for experiment: \n------ \n \n")
f.write("Number of filters: " + str([net1.in_filters]) + "\n")
f.write("Filter lengths: " + str([net1.filter_lengths]) + "\n")
f.write("Latend variables: " + str(dim_z) + "\n")
f.write("Iterations: " + str(iterations) + "\n")
f.write("Sample rate: " + str(sample_rate) + "\n")
f.close()
#device_play(net,sample_rate,duration = 1000)
#use np.squeeze to remove redundant dimentions.
#out = out.reshape(out.shape[0],out.shape[1],out.shape[3])
#print out2.shape
wav.write("sound/"+experiment_name+"_reconstruction_original.wav",sample_rate,original)
# Weights
plot_params(net.params,weight_plot_dir)
#ou1 = net1.output(x_train[:50])
# Reconstruction images
ou2 = net.output(x_train[:50])
for i in range(ou2.shape[0]):
if i <100:
plt.figure()
# plt.plot(ou1[i,0,:,0],color = "g")
plt.plot(ou2[i,0,:,0],color = "b")
plt.plot(x_train[i,0,:,0],color = "r")
plt.savefig(plot_dir+str(i)+"_compare.png",cmap=plt.cm.binary)
# The model
to_save = {"architecture":net_args,"weights":net.params,"dataset":data_name,"sample_rate":sample_rate}
pickle_saver(to_save,model_dir+experiment_name+"_model.pkl")
# The model hyper parameters
f = open(experiment_name+"readme","w")
f.write("SingVAE parameters for experiment: \n------ \n \n")
f.write("Number of filters: "+ str([net.in_filters]) +"\n")
f.write("Filter lengths: "+str([net.filter_lengths])+"\n")
f.write("Latend variables: "+str(net_args["dim_z"])+"\n")
f.write("Iterations: "+str(iterations)+"\n")
f.write("Sample rate: "+str(sample_rate)+"\n")
f.close()
wav.write("sound/"+experiment_name+"_reconstruction.wav",sample_rate,reconstruction)
wav.write("sound/"+experiment_name+"_reconstruction_original.wav",sample_rate,original)
# Weights
plot_params(net2.params,weight_plot_dir)
#ou1 = net1.output(x_train[:50])
# Reconstruction images
ou2 = net2.output(x_train[:50])
for i in range(ou2.shape[0]):
if i <100:
plt.figure()
# plt.plot(ou1[i,0,:,0],color = "g")
plt.plot(ou2[i,0,:,0],color = "b")
plt.plot(x_train[i,0,:,0],color = "r")
plt.savefig(plot_dir+str(i)+"_compare.png",cmap=plt.cm.binary)
# The model
pickle_saver(net2.params,model_dir+"model.pkl")
# The model hyper parameters
f = open("readme","w")
f.write("SingVAE parameters for experiment: \n------ \n \n")
f.write("Number of filters: "+ str([net1.in_filters]) +"\n")
f.write("Filter lengths: "+str([net1.filter_lengths])+"\n")
f.write("Latend variables: "+str(dim_z)+"\n")
f.write("Iterations: "+str(iterations)+"\n")
f.write("Sample rate: "+str(sample_rate)+"\n")
f.close()
print "Creating Theano functions"
encoder.createGradientFunctions()
print "Initializing weights and biases"
encoder.initParams()
lowerbound = np.array([])
testlowerbound = np.array([])
begin = time.time()
# Declare figures for interactive mode
plt.ion()
f1 = plt.figure(1)
for j in xrange(n_steps):
encoder.lowerbound = 0
print 'Iteration:', j
begin = time.time()
encoder.iterate(data)
end = time.time()
print("Iteration %d, lower bound = %.2f,"
" time = %.2fs"
% (j, encoder.lowerbound/N, end - begin))
if j%10 == 0:
mu_out3 = encoder.getTestOutput(data)
data.shape
plot_reconstructed(data[:10,:1000],mu_out3.T[:10,:1000],f1,interactive=True)
pickle_saver(encoder,"encoder.pkl")
random,reconstruction,original = sound_write.sample_write(net1,880,duration = 10)
wav.write("sound/"+experiment_name+"_random.wav",sample_rate,random)
wav.write("sound/"+experiment_name+"_reconstruction.wav",sample_rate,reconstruction)
wav.write("sound/"+experiment_name+"_reconstruction_original.wav",sample_rate,original)
ou1 = net1.output(x_train[:50])
ou2 = net2.output(x_train[:50])
for i in range(ou2.shape[0]):
if i <100:
plt.figure()
plt.plot(ou1[i,0,:,0],color = "g")
plt.plot(ou2[i,0,:,0],color = "b")
plt.plot(x_train[i,0,:,0],color = "r")
plt.savefig(plot_dir+str(i)+"_compare.png",cmap=plt.cm.binary)
pickle_saver(net2.params,"model_no_boost.pkl")
f = open("readme","w")
f.write("SingVAE parameters for experiment: \n------ \n \n")
f.write("Number of filters: "+ str([net1.in_filters]) +"\n")
f.write("Filter lengths: "+str([net1.filter_lengths])+"\n")
f.write("Latend variables: "+str(dim_z)+"\n")
f.write("Iterations: "+str(iterations)+"\n")
f.write("Sample rate: "+str(sample_rate)+"\n")
f.close()
#device_play(net,sample_rate,duration = 1000)
#use np.squeeze to remove redundant dimentions.
#out = out.reshape(out.shape[0],out.shape[1],out.shape[3])
#print out2.shape